To support more color channels, and greater flexibility in our rendering
engine, this new code impliments a new API for accessing pixel data in a
cairo surface and expands it's capabilities so that they can work with
floating point pixel data as well as CMYKA (5 channel) interpolations.

The new interface makes heavy use of compile time templating for channel count
and cairo surface formats.

The primary use of this API is for filters. Each filter which depends on
non-cairo functionality is recoded to use this new API and split out from
their svg filter primitive and display object files so they can be unit tested.

Threading and dispatch pool are held as light copies until the refactoring is
completed and we can copy them to a central utility instead.

Assistance from PBS (squashed):

* Use requires instead of run-time checks.
* Use more modern _v/_t instead of ::value/::type.
* Fix const T* memory() calling itself in infinite loop.
* Make some functions static (_primary_pos, _bilinear_interpolate).
* Make _transform std::optional.
* Convert G_BYTE_ORDER == G_LITTLE_ENDIAN to if constexpr, revealing
  missing glib.h header.
* Use safemod for wrapping.

To contain the multiple cairo surfaces and integrate with the pixel access code
a new Surface class is made that stores references to cairo surfaces.

PixelAccess objects of the right format can be run with filters to control the
surface at a pixel by pixel level.

A host of new tests are adding for each of the filters to make sure PixelAccess
integration is correct for each one of them.

Other Changes

 * Add forward declaration to colors API to smooth importing

We wrap the cairo context with our own multi-context engine
which keeps track of the color space, can paint multi-surface
CMYK rasters onto each other as well as translate between
Inkscape geom, path objects to cairo instructions.

Added many tests for each of the functionalities in cairo context
drawing.

Add README to explain the local code structures.

The stack mostly backs onto PixelFilters, but also onto the Surface
and Context where upstream functionality can provide the output.

The ColorSpace for the filter primitives are tied into the rendering
process so filters are rendered in the correct space.

Drawing options are refactored to be more concise for future use.

This copy from display allows the differences to be seen more easily.

display/drawing* -> renderer/drawing/drawing*
display/surface* -> renderer/drawing/drawing-area* (might not be needed)
display/tags* -> renderer/drawing/drawing-item-tags*
display/nr-style* -> renderer/drawing/drawing-style*

The process used for this refactoring was a light touch on the logic. But
the namespaces, and somme processes that were incompatible with the new
surface and pixel filters had to be changed.

The RenderContext was changed to DrawingOptions to better express what it is
and the various options from filters and other parts of the pipeline were
added so they aren't being passed around in addition.

Pull in all uses of SPStyle into the DrawingStyle parsing and remove DrawingItem
style variables to provide consistancy and remove duplication.

This tracks the traditional rendering display tree and prints out
new rendering engine c++ code which can be used as tests. Combined
with the make-test.sh command this can turn any svg file into a
rendering unit test.

Currently: Any PNG export will turn on logging. TODO: Make it a CLI option

Exported tests come from the pdfoutput test suite.